2 * Copyright (C) 2014 Facebook. All rights reserved.
4 * This file is released under the GPL.
7 #include <linux/device-mapper.h>
9 #include <linux/module.h>
10 #include <linux/init.h>
11 #include <linux/blkdev.h>
12 #include <linux/bio.h>
13 #include <linux/dax.h>
14 #include <linux/slab.h>
15 #include <linux/kthread.h>
16 #include <linux/freezer.h>
17 #include <linux/uio.h>
19 #define DM_MSG_PREFIX "log-writes"
22 * This target will sequentially log all writes to the target device onto the
23 * log device. This is helpful for replaying writes to check for fs consistency
24 * at all times. This target provides a mechanism to mark specific events to
25 * check data at a later time. So for example you would:
29 * dmsetup message /dev/whatever mark mymark
32 * Then replay the log up to mymark and check the contents of the replay to
33 * verify it matches what was written.
35 * We log writes only after they have been flushed, this makes the log describe
36 * close to the order in which the data hits the actual disk, not its cache. So
37 * for example the following sequence (W means write, C means complete)
39 * Wa,Wb,Wc,Cc,Ca,FLUSH,FUAd,Cb,CFLUSH,CFUAd
41 * Would result in the log looking like this:
43 * c,a,flush,fuad,b,<other writes>,<next flush>
45 * This is meant to help expose problems where file systems do not properly wait
46 * on data being written before invoking a FLUSH. FUA bypasses cache so once it
47 * completes it is added to the log as it should be on disk.
49 * We treat DISCARDs as if they don't bypass cache so that they are logged in
50 * order of completion along with the normal writes. If we didn't do it this
51 * way we would process all the discards first and then write all the data, when
52 * in fact we want to do the data and the discard in the order that they
55 #define LOG_FLUSH_FLAG (1 << 0)
56 #define LOG_FUA_FLAG (1 << 1)
57 #define LOG_DISCARD_FLAG (1 << 2)
58 #define LOG_MARK_FLAG (1 << 3)
59 #define LOG_METADATA_FLAG (1 << 4)
61 #define WRITE_LOG_VERSION 1ULL
62 #define WRITE_LOG_MAGIC 0x6a736677736872ULL
65 * The disk format for this is braindead simple.
67 * At byte 0 we have our super, followed by the following sequence for
70 * [ 1 sector ][ entry->nr_sectors ]
71 * [log_write_entry][ data written ]
73 * The log_write_entry takes up a full sector so we can have arbitrary length
74 * marks and it leaves us room for extra content in the future.
78 * Basic info about the log for userspace.
80 struct log_write_super {
88 * sector - the sector we wrote.
89 * nr_sectors - the number of sectors we wrote.
90 * flags - flags for this log entry.
91 * data_len - the size of the data in this log entry, this is for private log
92 * entry stuff, the MARK data provided by userspace for example.
94 struct log_write_entry {
101 struct log_writes_c {
103 struct dm_dev *logdev;
108 atomic_t pending_blocks;
109 sector_t next_sector;
111 bool logging_enabled;
112 bool device_supports_discard;
113 spinlock_t blocks_lock;
114 struct list_head unflushed_blocks;
115 struct list_head logging_blocks;
116 wait_queue_head_t wait;
117 struct task_struct *log_kthread;
120 struct pending_block {
127 struct list_head list;
128 struct bio_vec vecs[0];
131 struct per_bio_data {
132 struct pending_block *block;
135 static inline sector_t bio_to_dev_sectors(struct log_writes_c *lc,
138 return sectors >> (lc->sectorshift - SECTOR_SHIFT);
141 static inline sector_t dev_to_bio_sectors(struct log_writes_c *lc,
144 return sectors << (lc->sectorshift - SECTOR_SHIFT);
147 static void put_pending_block(struct log_writes_c *lc)
149 if (atomic_dec_and_test(&lc->pending_blocks)) {
150 smp_mb__after_atomic();
151 if (waitqueue_active(&lc->wait))
156 static void put_io_block(struct log_writes_c *lc)
158 if (atomic_dec_and_test(&lc->io_blocks)) {
159 smp_mb__after_atomic();
160 if (waitqueue_active(&lc->wait))
165 static void log_end_io(struct bio *bio)
167 struct log_writes_c *lc = bio->bi_private;
169 if (bio->bi_status) {
172 DMERR("Error writing log block, error=%d", bio->bi_status);
173 spin_lock_irqsave(&lc->blocks_lock, flags);
174 lc->logging_enabled = false;
175 spin_unlock_irqrestore(&lc->blocks_lock, flags);
184 * Meant to be called if there is an error, it will free all the pages
185 * associated with the block.
187 static void free_pending_block(struct log_writes_c *lc,
188 struct pending_block *block)
192 for (i = 0; i < block->vec_cnt; i++) {
193 if (block->vecs[i].bv_page)
194 __free_page(block->vecs[i].bv_page);
198 put_pending_block(lc);
201 static int write_metadata(struct log_writes_c *lc, void *entry,
202 size_t entrylen, void *data, size_t datalen,
210 bio = bio_alloc(GFP_KERNEL, 1);
212 DMERR("Couldn't alloc log bio");
215 bio->bi_iter.bi_size = 0;
216 bio->bi_iter.bi_sector = sector;
217 bio_set_dev(bio, lc->logdev->bdev);
218 bio->bi_end_io = log_end_io;
219 bio->bi_private = lc;
220 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
222 page = alloc_page(GFP_KERNEL);
224 DMERR("Couldn't alloc log page");
229 ptr = kmap_atomic(page);
230 memcpy(ptr, entry, entrylen);
232 memcpy(ptr + entrylen, data, datalen);
233 memset(ptr + entrylen + datalen, 0,
234 lc->sectorsize - entrylen - datalen);
237 ret = bio_add_page(bio, page, lc->sectorsize, 0);
238 if (ret != lc->sectorsize) {
239 DMERR("Couldn't add page to the log block");
252 static int write_inline_data(struct log_writes_c *lc, void *entry,
253 size_t entrylen, void *data, size_t datalen,
256 int num_pages, bio_pages, pg_datalen, pg_sectorlen, i;
263 num_pages = ALIGN(datalen, PAGE_SIZE) >> PAGE_SHIFT;
264 bio_pages = min(num_pages, BIO_MAX_PAGES);
266 atomic_inc(&lc->io_blocks);
268 bio = bio_alloc(GFP_KERNEL, bio_pages);
270 DMERR("Couldn't alloc inline data bio");
274 bio->bi_iter.bi_size = 0;
275 bio->bi_iter.bi_sector = sector;
276 bio_set_dev(bio, lc->logdev->bdev);
277 bio->bi_end_io = log_end_io;
278 bio->bi_private = lc;
279 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
281 for (i = 0; i < bio_pages; i++) {
282 pg_datalen = min_t(int, datalen, PAGE_SIZE);
283 pg_sectorlen = ALIGN(pg_datalen, lc->sectorsize);
285 page = alloc_page(GFP_KERNEL);
287 DMERR("Couldn't alloc inline data page");
291 ptr = kmap_atomic(page);
292 memcpy(ptr, data, pg_datalen);
293 if (pg_sectorlen > pg_datalen)
294 memset(ptr + pg_datalen, 0, pg_sectorlen - pg_datalen);
297 ret = bio_add_page(bio, page, pg_sectorlen, 0);
298 if (ret != pg_sectorlen) {
299 DMERR("Couldn't add page of inline data");
304 datalen -= pg_datalen;
309 sector += bio_pages * PAGE_SECTORS;
320 static int log_one_block(struct log_writes_c *lc,
321 struct pending_block *block, sector_t sector)
324 struct log_write_entry entry;
325 size_t metadatalen, ret;
328 entry.sector = cpu_to_le64(block->sector);
329 entry.nr_sectors = cpu_to_le64(block->nr_sectors);
330 entry.flags = cpu_to_le64(block->flags);
331 entry.data_len = cpu_to_le64(block->datalen);
333 metadatalen = (block->flags & LOG_MARK_FLAG) ? block->datalen : 0;
334 if (write_metadata(lc, &entry, sizeof(entry), block->data,
335 metadatalen, sector)) {
336 free_pending_block(lc, block);
340 sector += dev_to_bio_sectors(lc, 1);
342 if (block->datalen && metadatalen == 0) {
343 if (write_inline_data(lc, &entry, sizeof(entry), block->data,
344 block->datalen, sector)) {
345 free_pending_block(lc, block);
348 /* we don't support both inline data & bio data */
355 atomic_inc(&lc->io_blocks);
356 bio = bio_alloc(GFP_KERNEL, min(block->vec_cnt, BIO_MAX_PAGES));
358 DMERR("Couldn't alloc log bio");
361 bio->bi_iter.bi_size = 0;
362 bio->bi_iter.bi_sector = sector;
363 bio_set_dev(bio, lc->logdev->bdev);
364 bio->bi_end_io = log_end_io;
365 bio->bi_private = lc;
366 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
368 for (i = 0; i < block->vec_cnt; i++) {
370 * The page offset is always 0 because we allocate a new page
371 * for every bvec in the original bio for simplicity sake.
373 ret = bio_add_page(bio, block->vecs[i].bv_page,
374 block->vecs[i].bv_len, 0);
375 if (ret != block->vecs[i].bv_len) {
376 atomic_inc(&lc->io_blocks);
378 bio = bio_alloc(GFP_KERNEL, min(block->vec_cnt - i, BIO_MAX_PAGES));
380 DMERR("Couldn't alloc log bio");
383 bio->bi_iter.bi_size = 0;
384 bio->bi_iter.bi_sector = sector;
385 bio_set_dev(bio, lc->logdev->bdev);
386 bio->bi_end_io = log_end_io;
387 bio->bi_private = lc;
388 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
390 ret = bio_add_page(bio, block->vecs[i].bv_page,
391 block->vecs[i].bv_len, 0);
392 if (ret != block->vecs[i].bv_len) {
393 DMERR("Couldn't add page on new bio?");
398 sector += block->vecs[i].bv_len >> SECTOR_SHIFT;
404 put_pending_block(lc);
407 free_pending_block(lc, block);
412 static int log_super(struct log_writes_c *lc)
414 struct log_write_super super;
416 super.magic = cpu_to_le64(WRITE_LOG_MAGIC);
417 super.version = cpu_to_le64(WRITE_LOG_VERSION);
418 super.nr_entries = cpu_to_le64(lc->logged_entries);
419 super.sectorsize = cpu_to_le32(lc->sectorsize);
421 if (write_metadata(lc, &super, sizeof(super), NULL, 0, 0)) {
422 DMERR("Couldn't write super");
429 static inline sector_t logdev_last_sector(struct log_writes_c *lc)
431 return i_size_read(lc->logdev->bdev->bd_inode) >> SECTOR_SHIFT;
434 static int log_writes_kthread(void *arg)
436 struct log_writes_c *lc = (struct log_writes_c *)arg;
439 while (!kthread_should_stop()) {
441 bool logging_enabled;
442 struct pending_block *block = NULL;
445 spin_lock_irq(&lc->blocks_lock);
446 if (!list_empty(&lc->logging_blocks)) {
447 block = list_first_entry(&lc->logging_blocks,
448 struct pending_block, list);
449 list_del_init(&block->list);
450 if (!lc->logging_enabled)
453 sector = lc->next_sector;
454 if (!(block->flags & LOG_DISCARD_FLAG))
455 lc->next_sector += dev_to_bio_sectors(lc, block->nr_sectors);
456 lc->next_sector += dev_to_bio_sectors(lc, 1);
459 * Apparently the size of the device may not be known
460 * right away, so handle this properly.
463 lc->end_sector = logdev_last_sector(lc);
464 if (lc->end_sector &&
465 lc->next_sector >= lc->end_sector) {
466 DMERR("Ran out of space on the logdev");
467 lc->logging_enabled = false;
470 lc->logged_entries++;
471 atomic_inc(&lc->io_blocks);
473 super = (block->flags & (LOG_FUA_FLAG | LOG_MARK_FLAG));
475 atomic_inc(&lc->io_blocks);
478 logging_enabled = lc->logging_enabled;
479 spin_unlock_irq(&lc->blocks_lock);
481 if (logging_enabled) {
482 ret = log_one_block(lc, block, sector);
486 spin_lock_irq(&lc->blocks_lock);
487 lc->logging_enabled = false;
488 spin_unlock_irq(&lc->blocks_lock);
491 free_pending_block(lc, block);
495 if (!try_to_freeze()) {
496 set_current_state(TASK_INTERRUPTIBLE);
497 if (!kthread_should_stop() &&
498 list_empty(&lc->logging_blocks))
500 __set_current_state(TASK_RUNNING);
507 * Construct a log-writes mapping:
508 * log-writes <dev_path> <log_dev_path>
510 static int log_writes_ctr(struct dm_target *ti, unsigned int argc, char **argv)
512 struct log_writes_c *lc;
513 struct dm_arg_set as;
514 const char *devname, *logdevname;
521 ti->error = "Invalid argument count";
525 lc = kzalloc(sizeof(struct log_writes_c), GFP_KERNEL);
527 ti->error = "Cannot allocate context";
530 spin_lock_init(&lc->blocks_lock);
531 INIT_LIST_HEAD(&lc->unflushed_blocks);
532 INIT_LIST_HEAD(&lc->logging_blocks);
533 init_waitqueue_head(&lc->wait);
534 atomic_set(&lc->io_blocks, 0);
535 atomic_set(&lc->pending_blocks, 0);
537 devname = dm_shift_arg(&as);
538 ret = dm_get_device(ti, devname, dm_table_get_mode(ti->table), &lc->dev);
540 ti->error = "Device lookup failed";
544 logdevname = dm_shift_arg(&as);
545 ret = dm_get_device(ti, logdevname, dm_table_get_mode(ti->table),
548 ti->error = "Log device lookup failed";
549 dm_put_device(ti, lc->dev);
553 lc->sectorsize = bdev_logical_block_size(lc->dev->bdev);
554 lc->sectorshift = ilog2(lc->sectorsize);
555 lc->log_kthread = kthread_run(log_writes_kthread, lc, "log-write");
556 if (IS_ERR(lc->log_kthread)) {
557 ret = PTR_ERR(lc->log_kthread);
558 ti->error = "Couldn't alloc kthread";
559 dm_put_device(ti, lc->dev);
560 dm_put_device(ti, lc->logdev);
565 * next_sector is in 512b sectors to correspond to what bi_sector expects.
566 * The super starts at sector 0, and the next_sector is the next logical
567 * one based on the sectorsize of the device.
569 lc->next_sector = lc->sectorsize >> SECTOR_SHIFT;
570 lc->logging_enabled = true;
571 lc->end_sector = logdev_last_sector(lc);
572 lc->device_supports_discard = true;
574 ti->num_flush_bios = 1;
575 ti->flush_supported = true;
576 ti->num_discard_bios = 1;
577 ti->discards_supported = true;
578 ti->per_io_data_size = sizeof(struct per_bio_data);
587 static int log_mark(struct log_writes_c *lc, char *data)
589 struct pending_block *block;
590 size_t maxsize = lc->sectorsize - sizeof(struct log_write_entry);
592 block = kzalloc(sizeof(struct pending_block), GFP_KERNEL);
594 DMERR("Error allocating pending block");
598 block->data = kstrndup(data, maxsize - 1, GFP_KERNEL);
600 DMERR("Error copying mark data");
604 atomic_inc(&lc->pending_blocks);
605 block->datalen = strlen(block->data);
606 block->flags |= LOG_MARK_FLAG;
607 spin_lock_irq(&lc->blocks_lock);
608 list_add_tail(&block->list, &lc->logging_blocks);
609 spin_unlock_irq(&lc->blocks_lock);
610 wake_up_process(lc->log_kthread);
614 static int log_dax(struct log_writes_c *lc, sector_t sector, size_t bytes,
617 struct pending_block *block;
622 block = kzalloc(sizeof(struct pending_block), GFP_KERNEL);
624 DMERR("Error allocating dax pending block");
628 block->data = kzalloc(bytes, GFP_KERNEL);
630 DMERR("Error allocating dax data space");
635 /* write data provided via the iterator */
636 if (!copy_from_iter(block->data, bytes, i)) {
637 DMERR("Error copying dax data");
643 /* rewind the iterator so that the block driver can use it */
644 iov_iter_revert(i, bytes);
646 block->datalen = bytes;
647 block->sector = bio_to_dev_sectors(lc, sector);
648 block->nr_sectors = ALIGN(bytes, lc->sectorsize) >> lc->sectorshift;
650 atomic_inc(&lc->pending_blocks);
651 spin_lock_irq(&lc->blocks_lock);
652 list_add_tail(&block->list, &lc->unflushed_blocks);
653 spin_unlock_irq(&lc->blocks_lock);
654 wake_up_process(lc->log_kthread);
659 static void log_writes_dtr(struct dm_target *ti)
661 struct log_writes_c *lc = ti->private;
663 spin_lock_irq(&lc->blocks_lock);
664 list_splice_init(&lc->unflushed_blocks, &lc->logging_blocks);
665 spin_unlock_irq(&lc->blocks_lock);
668 * This is just nice to have since it'll update the super to include the
669 * unflushed blocks, if it fails we don't really care.
671 log_mark(lc, "dm-log-writes-end");
672 wake_up_process(lc->log_kthread);
673 wait_event(lc->wait, !atomic_read(&lc->io_blocks) &&
674 !atomic_read(&lc->pending_blocks));
675 kthread_stop(lc->log_kthread);
677 WARN_ON(!list_empty(&lc->logging_blocks));
678 WARN_ON(!list_empty(&lc->unflushed_blocks));
679 dm_put_device(ti, lc->dev);
680 dm_put_device(ti, lc->logdev);
684 static void normal_map_bio(struct dm_target *ti, struct bio *bio)
686 struct log_writes_c *lc = ti->private;
688 bio_set_dev(bio, lc->dev->bdev);
691 static int log_writes_map(struct dm_target *ti, struct bio *bio)
693 struct log_writes_c *lc = ti->private;
694 struct per_bio_data *pb = dm_per_bio_data(bio, sizeof(struct per_bio_data));
695 struct pending_block *block;
696 struct bvec_iter iter;
700 bool flush_bio = (bio->bi_opf & REQ_PREFLUSH);
701 bool fua_bio = (bio->bi_opf & REQ_FUA);
702 bool discard_bio = (bio_op(bio) == REQ_OP_DISCARD);
703 bool meta_bio = (bio->bi_opf & REQ_META);
707 /* Don't bother doing anything if logging has been disabled */
708 if (!lc->logging_enabled)
712 * Map reads as normal.
714 if (bio_data_dir(bio) == READ)
717 /* No sectors and not a flush? Don't care */
718 if (!bio_sectors(bio) && !flush_bio)
722 * Discards will have bi_size set but there's no actual data, so just
723 * allocate the size of the pending block.
726 alloc_size = sizeof(struct pending_block);
728 alloc_size = sizeof(struct pending_block) + sizeof(struct bio_vec) * bio_segments(bio);
730 block = kzalloc(alloc_size, GFP_NOIO);
732 DMERR("Error allocating pending block");
733 spin_lock_irq(&lc->blocks_lock);
734 lc->logging_enabled = false;
735 spin_unlock_irq(&lc->blocks_lock);
736 return DM_MAPIO_KILL;
738 INIT_LIST_HEAD(&block->list);
740 atomic_inc(&lc->pending_blocks);
743 block->flags |= LOG_FLUSH_FLAG;
745 block->flags |= LOG_FUA_FLAG;
747 block->flags |= LOG_DISCARD_FLAG;
749 block->flags |= LOG_METADATA_FLAG;
751 block->sector = bio_to_dev_sectors(lc, bio->bi_iter.bi_sector);
752 block->nr_sectors = bio_to_dev_sectors(lc, bio_sectors(bio));
754 /* We don't need the data, just submit */
756 WARN_ON(flush_bio || fua_bio);
757 if (lc->device_supports_discard)
760 return DM_MAPIO_SUBMITTED;
763 /* Flush bio, splice the unflushed blocks onto this list and submit */
764 if (flush_bio && !bio_sectors(bio)) {
765 spin_lock_irq(&lc->blocks_lock);
766 list_splice_init(&lc->unflushed_blocks, &block->list);
767 spin_unlock_irq(&lc->blocks_lock);
772 * We will write this bio somewhere else way later so we need to copy
773 * the actual contents into new pages so we know the data will always be
776 * We do this because this could be a bio from O_DIRECT in which case we
777 * can't just hold onto the page until some later point, we have to
778 * manually copy the contents.
780 bio_for_each_segment(bv, bio, iter) {
784 page = alloc_page(GFP_NOIO);
786 DMERR("Error allocing page");
787 free_pending_block(lc, block);
788 spin_lock_irq(&lc->blocks_lock);
789 lc->logging_enabled = false;
790 spin_unlock_irq(&lc->blocks_lock);
791 return DM_MAPIO_KILL;
794 src = kmap_atomic(bv.bv_page);
795 dst = kmap_atomic(page);
796 memcpy(dst, src + bv.bv_offset, bv.bv_len);
799 block->vecs[i].bv_page = page;
800 block->vecs[i].bv_len = bv.bv_len;
805 /* Had a flush with data in it, weird */
807 spin_lock_irq(&lc->blocks_lock);
808 list_splice_init(&lc->unflushed_blocks, &block->list);
809 spin_unlock_irq(&lc->blocks_lock);
812 normal_map_bio(ti, bio);
813 return DM_MAPIO_REMAPPED;
816 static int normal_end_io(struct dm_target *ti, struct bio *bio,
819 struct log_writes_c *lc = ti->private;
820 struct per_bio_data *pb = dm_per_bio_data(bio, sizeof(struct per_bio_data));
822 if (bio_data_dir(bio) == WRITE && pb->block) {
823 struct pending_block *block = pb->block;
826 spin_lock_irqsave(&lc->blocks_lock, flags);
827 if (block->flags & LOG_FLUSH_FLAG) {
828 list_splice_tail_init(&block->list, &lc->logging_blocks);
829 list_add_tail(&block->list, &lc->logging_blocks);
830 wake_up_process(lc->log_kthread);
831 } else if (block->flags & LOG_FUA_FLAG) {
832 list_add_tail(&block->list, &lc->logging_blocks);
833 wake_up_process(lc->log_kthread);
835 list_add_tail(&block->list, &lc->unflushed_blocks);
836 spin_unlock_irqrestore(&lc->blocks_lock, flags);
839 return DM_ENDIO_DONE;
843 * INFO format: <logged entries> <highest allocated sector>
845 static void log_writes_status(struct dm_target *ti, status_type_t type,
846 unsigned status_flags, char *result,
850 struct log_writes_c *lc = ti->private;
853 case STATUSTYPE_INFO:
854 DMEMIT("%llu %llu", lc->logged_entries,
855 (unsigned long long)lc->next_sector - 1);
856 if (!lc->logging_enabled)
857 DMEMIT(" logging_disabled");
860 case STATUSTYPE_TABLE:
861 DMEMIT("%s %s", lc->dev->name, lc->logdev->name);
866 static int log_writes_prepare_ioctl(struct dm_target *ti,
867 struct block_device **bdev, fmode_t *mode)
869 struct log_writes_c *lc = ti->private;
870 struct dm_dev *dev = lc->dev;
874 * Only pass ioctls through if the device sizes match exactly.
876 if (ti->len != i_size_read(dev->bdev->bd_inode) >> SECTOR_SHIFT)
881 static int log_writes_iterate_devices(struct dm_target *ti,
882 iterate_devices_callout_fn fn,
885 struct log_writes_c *lc = ti->private;
887 return fn(ti, lc->dev, 0, ti->len, data);
891 * Messages supported:
892 * mark <mark data> - specify the marked data.
894 static int log_writes_message(struct dm_target *ti, unsigned argc, char **argv,
895 char *result, unsigned maxlen)
898 struct log_writes_c *lc = ti->private;
901 DMWARN("Invalid log-writes message arguments, expect 2 arguments, got %d", argc);
905 if (!strcasecmp(argv[0], "mark"))
906 r = log_mark(lc, argv[1]);
908 DMWARN("Unrecognised log writes target message received: %s", argv[0]);
913 static void log_writes_io_hints(struct dm_target *ti, struct queue_limits *limits)
915 struct log_writes_c *lc = ti->private;
916 struct request_queue *q = bdev_get_queue(lc->dev->bdev);
918 if (!q || !blk_queue_discard(q)) {
919 lc->device_supports_discard = false;
920 limits->discard_granularity = lc->sectorsize;
921 limits->max_discard_sectors = (UINT_MAX >> SECTOR_SHIFT);
923 limits->logical_block_size = bdev_logical_block_size(lc->dev->bdev);
924 limits->physical_block_size = bdev_physical_block_size(lc->dev->bdev);
925 limits->io_min = limits->physical_block_size;
928 static long log_writes_dax_direct_access(struct dm_target *ti, pgoff_t pgoff,
929 long nr_pages, void **kaddr, pfn_t *pfn)
931 struct log_writes_c *lc = ti->private;
932 sector_t sector = pgoff * PAGE_SECTORS;
935 ret = bdev_dax_pgoff(lc->dev->bdev, sector, nr_pages * PAGE_SIZE, &pgoff);
938 return dax_direct_access(lc->dev->dax_dev, pgoff, nr_pages, kaddr, pfn);
941 static size_t log_writes_dax_copy_from_iter(struct dm_target *ti,
942 pgoff_t pgoff, void *addr, size_t bytes,
945 struct log_writes_c *lc = ti->private;
946 sector_t sector = pgoff * PAGE_SECTORS;
949 if (bdev_dax_pgoff(lc->dev->bdev, sector, ALIGN(bytes, PAGE_SIZE), &pgoff))
952 /* Don't bother doing anything if logging has been disabled */
953 if (!lc->logging_enabled)
956 err = log_dax(lc, sector, bytes, i);
958 DMWARN("Error %d logging DAX write", err);
962 return dax_copy_from_iter(lc->dev->dax_dev, pgoff, addr, bytes, i);
965 static struct target_type log_writes_target = {
966 .name = "log-writes",
967 .version = {1, 1, 0},
968 .module = THIS_MODULE,
969 .ctr = log_writes_ctr,
970 .dtr = log_writes_dtr,
971 .map = log_writes_map,
972 .end_io = normal_end_io,
973 .status = log_writes_status,
974 .prepare_ioctl = log_writes_prepare_ioctl,
975 .message = log_writes_message,
976 .iterate_devices = log_writes_iterate_devices,
977 .io_hints = log_writes_io_hints,
978 .direct_access = log_writes_dax_direct_access,
979 .dax_copy_from_iter = log_writes_dax_copy_from_iter,
982 static int __init dm_log_writes_init(void)
984 int r = dm_register_target(&log_writes_target);
987 DMERR("register failed %d", r);
992 static void __exit dm_log_writes_exit(void)
994 dm_unregister_target(&log_writes_target);
997 module_init(dm_log_writes_init);
998 module_exit(dm_log_writes_exit);
1000 MODULE_DESCRIPTION(DM_NAME " log writes target");
1001 MODULE_AUTHOR("Josef Bacik <jbacik@fb.com>");
1002 MODULE_LICENSE("GPL");